cleanup guest process when vm evacuate failed and then deleted

[Wenping Song]

when the nova_compute service is stopped and evacuate the vm on the host, then delete the vm when it is evacuating, the vm is local deleted by nova-api, but the libvirt process is not terminated, and the resources like GPU are still used by the vm process. from the code we can see when the host is init and the evacuation status is failed, this filter[1] cannot get the vm and the guest process is not destroy.[1]https://review.opendev.org/plugins/gitiles/openstack/nova/+/refs/heads/master/nova/compute/manager.py#703

[OpenStack Infra (hudson-openstack)]

Related fix proposed to branch: master
Review: https://review.opendev.org/c/openstack/nova/+/814746

[Sylvain Bauza (sylvain-bauza)]

Not sure I understand the issue.
If a compute service is not running, then evacuated instances are indeed moved but they are still in the compute service, sure. Then, when restarting the compute service, indeed we look at the evacuated instances from this host and then we delete the guests by [1] as you said.

Why do you want to delete also guests that were not evacuated ? That would mean that if a operators wanted to evacuate some instance but got an issue, we would delete the guest when restarting the nova-compute service while the operator wasn't able to move the instance before.

[1] https://github.com/openstack/nova/blob/00452a403b57723b364477082ce1587a909b2a6b/nova/compute/manager.py#L1440

Putting the bug report to Incomplete, but please put it back to New once you reply.

[Wenping Song (wenping1)]

the delete vm api doesnot intecept the evacuating vm, so the end use may delete while the vm is evacuating.
anyway, what will we do if the vm evacuates failed even we donot delete it during the vm is evacuating.